Testing prediction accuracy of hdu admission following high grade serous advanced ovarian cancer cytoreductive surgery using machine learning methods

Abstract

Introduction/Background Advanced high grade serous ovarian cancer patients (HGSOC) frequently require extensive procedures including bowel resections and upper abdominal surgery potentially necessitating HDU/ICU support and prolonged hospitalisation. HDU/ICU admission is a measurable outcome that can be used as a benchmark of surgical care. Modern data mining technologies such as Machine Learning (ML), a subfield of Artificial Intelligence, could be helpful in monitoring HDU/ICU admissions to improve standards of care. We aimed to improve the accuracy of predicting HDU admission in that cohort of patients by use of ML algorithms.

Methodology A cohort of 176 HGSOC patients, who underwent surgical cytoreduction from Jan 2014 to Dec 2017 was selected from the ovarian database. They were randomly assigned to ‘training’ and ‘test’ subcohorts. ML methods including Classification and Regression Trees (CART) and Support Vector Machine (SVM), were employed to derive predictive information for HDU/ICU admission from a list of selected preoperative, intraoperative, and postoperative variables. These methods were tested against conventional linear regression analyses.

Results There were 29 out of 176 (16.4%) HDU/ICU admissions; 23 admissions were elective whilst six were unplanned admissions. For the outcome of HDU/ICU admission, both ML methods outperformed conventional regression by far (table 1). Bowel resection and operative time were the most predictive variables (figure 1). HDU/ICU admission was not associated with increased length of stay, increased number of postoperative complications, and increased risk of readmission within 30 days.